Practitioners employ instance-wise feature selection algorithms to understand their data at the granularity of a single sample. Providing explanations for each sample in a dataset can be computationally expensive. Recent methods amortize this cost by learning a selector model that takes a sample of data and identifies the subset of its features that is important. This subset gets passed to a predictor model for the target, which is learned in concert with the selector model to fit the data. We s…